It is common for most projects to use only one method from the OpenAI JavaScript SDK’s implementation, so why not use native fetch for the this one REST endpoint instead?

Question?

In AI web projects or any web AI template, if you want to access OpenAI API, why are you using openai npm module rather than the REST API endpoints directly?

I personally think using REST endpoints is the best possible approach in terms of runtimes compatibility.

For JavaScript AI projects the default recommendation should be to use native fetch and move to SDK later if necessary.

Performance benefits are clear:

• Reduced bundle size
• Using native fetch for compatibility with any runtime (edge, workers, deno, browser)
• streaming

  const res = await fetch('https://api.openai.com/v1/chat/completions', {
    headers: requestHeaders,
    method: 'POST',
    body: JSON.stringify(payload),
  })

If you are running long-running AI calls, remember to use Edge Functions instead of Serverless Functions:

References:

• Vercel Chat bot example using native fetch
• link to code snippet example

Godspeed

PS. Follow me on Twitter or LinkedIn
https://twitter.com/dom_sipowicz
https://www.linkedin.com/in/dominiksipowicz/